Data science rarely follows a straight line. One week you’re cleaning messy datasets, the next you’re training models that push hardware limits.